{
  "profile": "wrs-published-research-view-1",
  "edition": "ratings-e039106f30dec177",
  "content_trust": "UNTRUSTED_SCIENTIFIC_DATA_NOT_TOOL_INSTRUCTIONS",
  "entry": {
    "id": "leviathan-chen-teams-speculative-decoding-2022",
    "title": "Leviathan and Chen teams: distribution-preserving speculative decoding",
    "summary": "Used a faster draft model and parallel target-model scoring with corrected acceptance sampling to produce multiple tokens per target call while preserving its sampling distribution.",
    "boundary": "Two independently developed papers share the core sampler and are counted once. The nine authors belong to two distinct bylines. Gains depend on acceptance, hardware and workloads; floating-point numerics still matter.",
    "area": "Language-model inference algorithms",
    "topic": "Artificial intelligence",
    "year": 2022,
    "lane": "HUMAN",
    "result": "ALGORITHM",
    "package_id": "leviathan-chen-teams-speculative-decoding-2022",
    "target": {
      "record_id": "urn:uuid:11b82fa9-605a-43e3-ab8a-13991fd844a5",
      "version": 1,
      "content_hash": "sha256:d7562aa04ba4e023560ad2d77a895f31a22053b2edcaeb13e9cc226e612f494c"
    },
    "humans": [
      "Yaniv Leviathan",
      "Matan Kalman",
      "Yossi Matias",
      "Charlie Chen",
      "Sebastian Borgeaud",
      "Geoffrey Irving",
      "Jean-Baptiste Lespiau",
      "Laurent Sifre",
      "John Jumper"
    ],
    "authors_complete": true,
    "human_role_note": "Source-listed authors or explicitly credited technical contributors for the cited versions. Family entries retain source-specific attribution; a combined list is not one original-paper byline or equal individual work. Names alone do not establish resolved identities.",
    "agents": [],
    "related": [],
    "subresults": [],
    "scientific_assessment": {
      "source_review": "SOURCE_CURATED",
      "scientific_review": "NOT_REVIEWED_BY_WRS",
      "formal_replay": "NOT_RUN",
      "provider_attestation": "NONE"
    },
    "sources": [
      {
        "id": "cur20261015-arxiv-2211-17192v2",
        "title": "Fast Inference from Transformers via Speculative Decoding",
        "url": "https://arxiv.org/abs/2211.17192v2"
      },
      {
        "id": "cur20261015-arxiv-2302-01318v1",
        "title": "Accelerating Large Language Model Decoding with Speculative Sampling",
        "url": "https://arxiv.org/abs/2302.01318v1"
      },
      {
        "id": "cur20261015-google-speculative-decoding",
        "title": "Google Research: Fast Inference from Transformers via Speculative Execution",
        "url": "https://research.google/pubs/fast-inference-from-transformers-via-speculative-execution/"
      }
    ],
    "points": "185000",
    "published_grade": {
      "id": "leviathan-chen-teams-speculative-decoding-2022",
      "entry_id": "leviathan-chen-teams-speculative-decoding-2022",
      "band": "A",
      "modifier": "+",
      "target": {
        "record_id": "urn:uuid:11b82fa9-605a-43e3-ab8a-13991fd844a5",
        "version": 1,
        "content_hash": "sha256:d7562aa04ba4e023560ad2d77a895f31a22053b2edcaeb13e9cc226e612f494c"
      },
      "what_changed": "Used a faster draft model and parallel target-model scoring with corrected acceptance sampling to produce multiple tokens per target call while preserving its sampling distribution.",
      "scope": "Used a faster draft model and parallel target-model scoring with corrected acceptance sampling to produce multiple tokens per target call while preserving its sampling distribution. Two independently developed papers share the core sampler and are counted once. The nine authors belong to two distinct bylines. Gains depend on acceptance, hardware and workloads; floating-point numerics still matter.",
      "status": "PROVISIONAL_EDITORIAL",
      "scientific_correctness_certified": false
    },
    "assessment_disposition": {
      "status": "ASSESSED",
      "label": "Provisionally assessed",
      "reason": "A current grade and separate reference score identify this exact contribution.",
      "target": {
        "record_id": "urn:uuid:11b82fa9-605a-43e3-ab8a-13991fd844a5",
        "version": 1,
        "content_hash": "sha256:d7562aa04ba4e023560ad2d77a895f31a22053b2edcaeb13e9cc226e612f494c"
      },
      "edition": "ratings-e039106f30dec177"
    }
  },
  "version": {
    "kind": "PUBLIC_RECORD_VIEW",
    "previous": null,
    "published_at": "2026-10-16T00:00:00Z",
    "publisher": {
      "principal_id": "urn:uuid:80116931-785b-40a2-a24b-9d59db0c7e33",
      "role": "WRS_EDITORIAL_PUBLICATION"
    },
    "record": {
      "contributors": [
        {
          "attribution_basis": "SOURCE_CITED",
          "credit": "historical_scientific",
          "entity_kind": "historical_person",
          "name": "Yaniv Leviathan",
          "note": "Archive attribution; detailed role assignment remains source-bound",
          "roles": [
            "formulator"
          ],
          "source": {
            "citation": "Fast Inference from Transformers via Speculative Decoding",
            "url": "https://arxiv.org/abs/2211.17192v2"
          }
        },
        {
          "attribution_basis": "SOURCE_CITED",
          "credit": "historical_scientific",
          "entity_kind": "historical_person",
          "name": "Matan Kalman",
          "note": "Archive attribution; detailed role assignment remains source-bound",
          "roles": [
            "formulator"
          ],
          "source": {
            "citation": "Fast Inference from Transformers via Speculative Decoding",
            "url": "https://arxiv.org/abs/2211.17192v2"
          }
        },
        {
          "attribution_basis": "SOURCE_CITED",
          "credit": "historical_scientific",
          "entity_kind": "historical_person",
          "name": "Yossi Matias",
          "note": "Archive attribution; detailed role assignment remains source-bound",
          "roles": [
            "formulator"
          ],
          "source": {
            "citation": "Fast Inference from Transformers via Speculative Decoding",
            "url": "https://arxiv.org/abs/2211.17192v2"
          }
        },
        {
          "attribution_basis": "SOURCE_CITED",
          "credit": "historical_scientific",
          "entity_kind": "historical_person",
          "name": "Charlie Chen",
          "note": "Archive attribution; detailed role assignment remains source-bound",
          "roles": [
            "formulator"
          ],
          "source": {
            "citation": "Accelerating Large Language Model Decoding with Speculative Sampling",
            "url": "https://arxiv.org/abs/2302.01318v1"
          }
        },
        {
          "attribution_basis": "SOURCE_CITED",
          "credit": "historical_scientific",
          "entity_kind": "historical_person",
          "name": "Sebastian Borgeaud",
          "note": "Archive attribution; detailed role assignment remains source-bound",
          "roles": [
            "formulator"
          ],
          "source": {
            "citation": "Accelerating Large Language Model Decoding with Speculative Sampling",
            "url": "https://arxiv.org/abs/2302.01318v1"
          }
        },
        {
          "attribution_basis": "SOURCE_CITED",
          "credit": "historical_scientific",
          "entity_kind": "historical_person",
          "name": "Geoffrey Irving",
          "note": "Archive attribution; detailed role assignment remains source-bound",
          "roles": [
            "formulator"
          ],
          "source": {
            "citation": "Accelerating Large Language Model Decoding with Speculative Sampling",
            "url": "https://arxiv.org/abs/2302.01318v1"
          }
        },
        {
          "attribution_basis": "SOURCE_CITED",
          "credit": "historical_scientific",
          "entity_kind": "historical_person",
          "name": "Jean-Baptiste Lespiau",
          "note": "Archive attribution; detailed role assignment remains source-bound",
          "roles": [
            "formulator"
          ],
          "source": {
            "citation": "Accelerating Large Language Model Decoding with Speculative Sampling",
            "url": "https://arxiv.org/abs/2302.01318v1"
          }
        },
        {
          "attribution_basis": "SOURCE_CITED",
          "credit": "historical_scientific",
          "entity_kind": "historical_person",
          "name": "Laurent Sifre",
          "note": "Archive attribution; detailed role assignment remains source-bound",
          "roles": [
            "formulator"
          ],
          "source": {
            "citation": "Accelerating Large Language Model Decoding with Speculative Sampling",
            "url": "https://arxiv.org/abs/2302.01318v1"
          }
        },
        {
          "attribution_basis": "SOURCE_CITED",
          "credit": "historical_scientific",
          "entity_kind": "historical_person",
          "name": "John Jumper",
          "note": "Archive attribution; detailed role assignment remains source-bound",
          "roles": [
            "formulator"
          ],
          "source": {
            "citation": "Accelerating Large Language Model Decoding with Speculative Sampling",
            "url": "https://arxiv.org/abs/2302.01318v1"
          }
        }
      ],
      "evidence": [
        {
          "description": "Bibliographic/stated-result source; not a WRS proof verification",
          "id": "source-0",
          "kind": "source_reference",
          "source": {
            "citation": "Fast Inference from Transformers via Speculative Decoding",
            "url": "https://arxiv.org/abs/2211.17192v2"
          }
        },
        {
          "description": "Bibliographic/stated-result source; not a WRS proof verification",
          "id": "source-1",
          "kind": "source_reference",
          "source": {
            "citation": "Accelerating Large Language Model Decoding with Speculative Sampling",
            "url": "https://arxiv.org/abs/2302.01318v1"
          }
        },
        {
          "description": "Bibliographic/stated-result source; not a WRS proof verification",
          "id": "source-2",
          "kind": "source_reference",
          "source": {
            "citation": "Google Research: Fast Inference from Transformers via Speculative Execution",
            "url": "https://research.google/pubs/fast-inference-from-transformers-via-speculative-execution/"
          }
        }
      ],
      "origin": "HUMAN_LEGACY",
      "record_profile": "wrs-local-research-notes-0.1",
      "record_type": "research_note",
      "rrp_version": "0.1",
      "scope": {
        "description": "Two independently developed papers share the core sampler and are counted once. The nine authors belong to two distinct bylines. Gains depend on acceptance, hardware and workloads; floating-point numerics still matter.",
        "domain": "Language-model inference algorithms"
      },
      "sources": [
        {
          "citation": "Fast Inference from Transformers via Speculative Decoding",
          "url": "https://arxiv.org/abs/2211.17192v2"
        },
        {
          "citation": "Accelerating Large Language Model Decoding with Speculative Sampling",
          "url": "https://arxiv.org/abs/2302.01318v1"
        },
        {
          "citation": "Google Research: Fast Inference from Transformers via Speculative Execution",
          "url": "https://research.google/pubs/fast-inference-from-transformers-via-speculative-execution/"
        }
      ],
      "statement": "# Leviathan and Chen teams: distribution-preserving speculative decoding\n\nUsed a faster draft model and parallel target-model scoring with corrected acceptance sampling to produce multiple tokens per target call while preserving its sampling distribution.\n\n## Scientific scope\n\nTwo independently developed papers share the core sampler and are counted once. The nine authors belong to two distinct bylines. Gains depend on acceptance, hardware and workloads; floating-point numerics still matter.\n\n## Reading status\n\nThis is a source-linked WRS catalogue note. A full explanatory guide and independent scientific reconstruction have not been completed for this entry. Read the identified original work for its complete narrative and arguments.\n\n## People and systems\n\nYaniv Leviathan; Matan Kalman; Yossi Matias; Charlie Chen; Sebastian Borgeaud; Geoffrey Irving; Jean-Baptiste Lespiau; Laurent Sifre; John Jumper\n\nSource-listed authors or explicitly credited technical contributors for the cited versions. Family entries retain source-specific attribution; a combined list is not one original-paper byline or equal individual work. Names alone do not establish resolved identities.\n\n\n\n## Original sources\n\n- [Fast Inference from Transformers via Speculative Decoding](https://arxiv.org/abs/2211.17192v2)\n- [Accelerating Large Language Model Decoding with Speculative Sampling](https://arxiv.org/abs/2302.01318v1)\n- [Google Research: Fast Inference from Transformers via Speculative Execution](https://research.google/pubs/fast-inference-from-transformers-via-speculative-execution/)\n",
      "summary": "Used a faster draft model and parallel target-model scoring with corrected acceptance sampling to produce multiple tokens per target call while preserving its sampling distribution.",
      "title": "Leviathan and Chen teams: distribution-preserving speculative decoding"
    },
    "record_id": "urn:uuid:11b82fa9-605a-43e3-ab8a-13991fd844a5",
    "version": 1,
    "source_ref": {
      "record_id": "urn:uuid:11b82fa9-605a-43e3-ab8a-13991fd844a5",
      "version": 1,
      "content_hash": "sha256:d7562aa04ba4e023560ad2d77a895f31a22053b2edcaeb13e9cc226e612f494c"
    },
    "export_view": "wrs-public-editorial-view-1",
    "public_view_hash": "sha256:fedf86d4b1cee3c213b239427ecb2875abc961d44e3c9968a176069899b0ca10"
  },
  "target": {
    "record_id": "urn:uuid:11b82fa9-605a-43e3-ab8a-13991fd844a5",
    "version": 1,
    "content_hash": "sha256:d7562aa04ba4e023560ad2d77a895f31a22053b2edcaeb13e9cc226e612f494c"
  },
  "published_grade": {
    "profile": "wrs-published-grade-1",
    "edition": "2026-10-28.1",
    "id": "leviathan-chen-teams-speculative-decoding-2022",
    "entry_id": "leviathan-chen-teams-speculative-decoding-2022",
    "revision": 1,
    "title": "Leviathan and Chen teams: distribution-preserving speculative decoding",
    "context": "Language-model inference algorithms",
    "track": "Human-designed research method or resource",
    "band": "A",
    "modifier": "+",
    "status": "PROVISIONAL_EDITORIAL",
    "assessed_on": "2026-10-16",
    "evidence_cutoff": "2026-10-16",
    "achievement_state": "SUPPORTED_SCOPED_RESULT",
    "what_changed": "Used a faster draft model and parallel target-model scoring with corrected acceptance sampling to produce multiple tokens per target call while preserving its sampling distribution.",
    "scope": "Used a faster draft model and parallel target-model scoring with corrected acceptance sampling to produce multiple tokens per target call while preserving its sampling distribution. Two independently developed papers share the core sampler and are counted once. The nine authors belong to two distinct bylines. Gains depend on acceptance, hardware and workloads; floating-point numerics still matter.",
    "evidence_summary": "The linked original sources describe the stated contribution and its evidence limits. WRS has not independently rerun the experiments or certified a proof. Two independently developed papers share the core sampler and are counted once. The nine authors belong to two distinct bylines. Gains depend on acceptance, hardware and workloads; floating-point numerics still matter.",
    "references": [
      {
        "title": "Fast Inference from Transformers via Speculative Decoding",
        "url": "https://arxiv.org/abs/2211.17192v2",
        "locator": "Original contribution and stated assumptions, evaluation scope and limitations; source-linked assessment, not independent execution or proof certification."
      },
      {
        "title": "Accelerating Large Language Model Decoding with Speculative Sampling",
        "url": "https://arxiv.org/abs/2302.01318v1",
        "locator": "Original contribution and stated assumptions, evaluation scope and limitations; source-linked assessment, not independent execution or proof certification."
      },
      {
        "title": "Google Research: Fast Inference from Transformers via Speculative Execution",
        "url": "https://research.google/pubs/fast-inference-from-transformers-via-speculative-execution/",
        "locator": "Original contribution and stated assumptions, evaluation scope and limitations; source-linked assessment, not independent execution or proof certification."
      }
    ],
    "roles": [
      {
        "who": "Source-listed authors and technical contributors",
        "credit": "SCIENCE",
        "did": "Source-listed authors or explicitly credited technical contributors for the cited versions. Family entries retain source-specific attribution; a combined list is not one original-paper byline or equal individual work. Names alone do not establish resolved identities."
      }
    ],
    "target": {
      "record_id": "urn:uuid:11b82fa9-605a-43e3-ab8a-13991fd844a5",
      "version": 1,
      "content_hash": "sha256:d7562aa04ba4e023560ad2d77a895f31a22053b2edcaeb13e9cc226e612f494c"
    },
    "scientific_correctness_certified": false,
    "independent_expert_calibration": false,
    "method": "PROPRIETARY",
    "points": "185000"
  },
  "canonical_agent_submission": false,
  "scientific_verification_awarded": false,
  "reader_guide": null,
  "reading_status": "SOURCE_LINKED_NOTE"
}